When an image of an object having a surface with bumps, such as a fruit, wood, or human skin, is captured by a low-quality image capturing device, or when the object has inadequate dimensions to be clearly image-captured, the bumps on the surface are often not shown in an image due to an insufficiency of resolution. Various solutions have been proposed to address this kind of problem. It is known from Patent Reference 1 to provide a method of (i) capturing a first image by a digital camera or the like and a higher-quality second image by zooming a portion of the first image, then (ii) learning a quality improvement function from a relationship between the first image and the second image, and (iii) applying the quality improvement function to the entire image. As a result, the entire image has higher quality and higher resolution. This method enables low-quality image capturing devices to generate an image with higher quality, reconstructing information of the bumps, which has been prevented from an insufficiency of resolution, using information generated from the second image.
Although the method disclosed in Patent Reference 1 achieves the image generation showing the bumps, the method fails image generation under a pseudo light source different from an actual light source. The image generation under a pseudo light source needs information regarding the surface structure of the object, more specifically, a geometric parameter regarding geometric normal of the object surface and a viewpoint. However, Patent Reference 1 does not disclose any method of estimating, from image data, a geometric parameter that is different from the image data. In the method of Patent Reference 1, the quality improvement function is assumed to be generated directly from the image data, although this method is not limited for image generation.
There is another method of generating a geometric parameter regarding a shape of an object using a range finder or the like, thereby obtaining a geometric parameter indicating a macro shape of the object. Unfortunately again, this method has problems in resolution and fineness. Reconstruction of fine-scale bumps on a surface of an object, such as fruit, wood, or human skin, needs highly complicated functions in a used device, which results in unrealistic size and cost especially in terms of usability.
On the other hand, Patent Reference 2 discloses a method capable of generating an image showing bumps on a surface of an object, and also capable of estimating, from image data, a geometric parameter that is not the image data itself but information regarding a geometric normal of the object surface. In the technique disclosed in Patent Reference 2, Torrance-Sparrow model indicating respective physical relationships between pixel values and geometric parameters regarding a viewpoint position, a light source position, and a geometric normal of the object surface is applied to an image of the object. Then, each difference between the result and an actual measurement value is modeled using Gaussian distribution. Here, the components modeled by Gaussian distribution can be considered as components having higher resolution. Therefore, addition of the components modeled by Gaussian distribution to an image without information of bumps on a surface of the object enables the image to show the bumps. Moreover, the use of the geometric parameter regarding a viewpoint position, a light source position, and a geometric normal of an object surface makes it possible to generate an image under a pseudo light source.    [Patent Reference 1] US Patent Application Publication No. 2006/0003328A1    [Patent Reference 2] Japanese Unexamined Patent Application Publication No. 2006-31595